This is the 52nd post in my Theories of Knowledge and Reality series. In my last post, I looked at Frank Jackson's argument for property dualism, concluding the major arguments involving dualism and materialism about the human mind. This last mind-related post covers artificial intelligence, particularly whether a computer program could be enough to generate genuine thinking.
The strongest argument that a computer might think is an
argument from analogy. At least some examples of artificial intelligence in
science fiction seem to do the same things we do when we think. In Star Trek: the Next Generation, Lt.
Commander Data is an android who certainly seems to be a conscious, thinking
being. In one episode, Starfleet conducted a trial to determine whether he was
the property of Starfleet or whether he has rights enough to refuse to be
dismantled for research into artificial intelligence. The argument that won the
day is that, though we can't prove Data to have a mind, we also can't prove
anyone besides ourselves to be consciously aware. They do the things we do when
we're consciously aware, and they have similar brain states according to our
best science, but is it absolute proof? So it doesn't seem like Data is in much
worse shape than any normal human being, right? We should at least give him the
benefit of the doubt when it comes to moral issues.
Nevertheless, John Searle's Chinese Room example is
designed to show that a computer program that appears to think isn't thereby
thinking. You might be able to design a program that follows steps to appear to
think, but that doesn't mean it really understands anything.
Put a man in a room and give him instructions about what to
do when he sees certain symbols. He is to follow the instructions and write
down different symbols as a result. Little does he know that the symbols he
receives are actual Chinese questions, and he's giving back actual Chinese
answers. From the outside, someone might think someone inside understands
Chinese and is answering the questions, but it's all based on rules. This is
exactly what people trying to develop artificial intelligence are trying to
accomplish. Searle says it's a misguided goal if people think this is genuine
thinking and understanding, since the Chinese Room case shows that no one is
thinking, even though all the behavioral responses to stimuli indicate that
someone must be thinking. This is a problem for functionalism, since all the
functional roles are present. It's a problem for the Artificial Intelligence
project, since something like this could be developed, but Searle insists that
merely accomplishing it doesn't give us anything that thinks.
Searle considers some objections. The Systems Reply
admits that the man in the room doesn't understand, and the room itself
certainly doesn't, nor does the instruction book. But the whole system - the
man, the room, and the instruction book - does understand Chinese. Searle
responds that he can make the system disappear by having the man memorize all
the rules. The room does little work here. Now give the man Chinese questions,
and he can write the proper answers. He acts as if he understands, yet there's
no way he does - and he is the system. So the system doesn't understand.
The Robot Reply says what's missing is involvement in
the world. Just language isn't enough. Make it interact with the world in more
ways by putting the program in a robot that can talk, move around, play catch,
etc. In that case, it seems more as if it thinks. A computer doesn't get to
hold things in its hand and move around. It has no contact with the things its statements
are about. Some have thought that giving it contact with those things would
make it easier to see it as understanding what the statements are about.
Searle gives two problems for this reply. First, it concedes
a bit much for the artificial intelligence thesis to be true anymore. Thinking
is no longer about symbol manipulation but has to be caused in the right way
and lead to certain kinds of behavior. It's not all based on just getting the
right program. Second, simply getting a machine to move around and interact
with the world doesn't make it think. Put a person inside the robot and give
her instructions as with the man in the Chinese Room, and you would get the
same result - it doesn't seem as if thinking is going on here.
There's an even easier reply that Searle could make (but
doesn't). He can go back to his example of the person inside the room becoming
the system. This is a person who moves around and can interact with things.
This person can even know that these statements are about these things somehow.
But that doesn't require the person to know which words mean what. I can know a
Chinese statement is about the apple in my hand without knowing what the words
mean. So interacting with the world can't be enough for me to understand what my
statements are about.
The Brain Simulator Reply says that what's missing in
the Chinese Room case is that it's not based enough on the actual human brain.
Base it more directly on how neurons cause neurons to do things and such, and
then maybe you'd be more inclined to call it genuine thinking. First of all,
Searle points out that the whole idea was to get the right program, and then it
thinks, regardless of the actual structure of the thing doing the thinking. It
no longer fits with this if you model it directly on the human brain. It's no
longer discovering the right program but is now just duplicating aspects of
human brains. Second, you can do something like this with a person inside.
Instead of manipulating symbols in a room, imagine that he has a complex system
of water pipes, and he manipulates levers and nozzles so that water moves
through pipes the way electrical signals do in the brain. Model it on the human
brain. There's still no understanding.
Finally, the Combination Reply says to take aspects
of the previous examples and combine them into one, so this would be an
interactive robot with a computerized brain modeled directly on a human brain
with behavior indistinguishable from human behavior, and then we'd be more
inclined to think the whole system thinks. Searle admits that we might find
that claim to be more reasonable. The problem is that now we're as far away
from stumbling on the right program as we could get. We haven't discovered a
program but have simply made something very close to a human. When we say apes
think and feel, it's because we can't find any other explanation of their
behavior, and they have a similar enough brain to ours. If we say that about
this robot, it's for the same reasons. If we discover that it's just a program,
we'd be inclined to say there's no thinking going on.
Searle insists that human thinking is based on the human
brain, and our minds are just our brains. He resists the idea that thinking
might occur apart from actual human brains. Any thinking must be based in
something very close to the human brain. Consider the seeming possibility of a human body acting purely according to physical laws but not actually experiencing anything (philosopher David Chalmers has coined a technical term in philosophy for such a being: he calls them Zombies). Or consider someone who experiences things differently from most people even while having the same brain state (philosophers call such people Mutants, again coining a technical term that doesn't coincide with normal usage). Zombies and Mutants, in these senses, are impossible, according to Searle. Something had better be close enough
to the normal human brain, or it doesn't have pain, boredom, or thoughts such
as the thought that 2 + 2 = 4. Searle just has to deny that Mutants are possible,
something David Lewis, who started out with a view similar to Searle's, didn't
want to insist on, since there's no real argument for it. Zombies also couldn't
exist, since something with a human brain automatically thinks. Maybe this is
right (many materialists besides Searle think so, e.g. Simon Blackburn), but
it's hard to prove such a thing.